Laila Shereen Sakr and Cathy Thomas
Thursday, April 30, 2026

Carsey Wolf-Center
Media Industries in AI Inititiave

A Contrapuntal Approach to Global Data 

Roadmap

PROBLEM

• Global data is uneven, multilingual, power-structured

• AI systems
  flatten context

• Result: loss of meaning, misrepresentation

METHOD

• Glitch → reading systems against the grain
• Contrapuntal → holding multiple histories in relation
• Culturally aware ontological approach

SYSTEM

• R-Shief (archive → analysis)
• Contrapuntal AI (design → intervention)

CONTRIBUTION

• New dataset design principles
• Humanities inside ML pipelines
• AI as knowledge infrastructure

Research Question

For example, sentiment analysis fails to recognize sarcastic humor (2012).

How can AI systems represent global data without flattening the historical, linguistic, and political relations that give it meaning?

Challenge
Global data is uneven, multilingual, power-structured and AI flattens context.

• 70+ languages

• billions of posts

• global archive of social media

R-Shief (2009+)

Data extraction was not enough

• Pattern ≠ meaning

• Scale ≠ interpretation

• Multilingual data ≠ contextual understanding

• Archive required new epistemology

What R-Shief Taught Us

What Contrapuntal Will

Give Us

• Human–AI collaborative platform

• Data Visualization

• Contextual metadata

• Citational lineage

• Participatory annotation

• Small language models

• Bias evaluation Tool

• Online Community

• Shared Governance

Methodology

1. Data Layer (historical and culturally based)

• contextual metadata

• multilingual structure

• historical annotation

2. Model Layer (linguistic based)

• small language models

• comparative training (baseline vs enriched datasets)

3. Evaluation Layer (community based)

• interpretability

• bias mitigation

• semantic fidelity across languages

Contrapuntal

• multiple histories

• in relation

• without resolution

Capitol Mosaic, 30-ft wall of Jan 6 images from Parler and Twitter, 
VJ Um Amel (Qualcomm Gallery, 2021).

Participatory Design

Collaborators

• Network of Arab Women in AI
  Co-Founder, along with colleagues in Egypt, Lebanon, Germany, Canada, and Qatar
• Autonomous Futures
  Co-Founder, along with colleagues at Michigan Central, Ohio State University, META, and Arizona State University
• CWC Media Industries and AI

Acknowledgements

Students

Denise Alfonso
Ripley Baker
Henry Coburn
Elisa Coccioli
Jessie Ding
Calista Dollag
Lexxus Edison Coffey
Kamaya Jackson
Jiyoo Kim-Jung
Corinna Kelley
Cass Mayeda
 

Funders

• HFA Small Grants
• Arnhold Collaborative Research Grant
• AGI Grant
   

Sofia Mosqueda

Anna Shaverdyan
Shashank Shivashankar
Saide Singh
Justus Wan
Calais Waring
Winston Zuo
 

Consultants:
Josh Bevan
Sierra Peltcher